As large language models like OpenAI's GPT-4 become more advanced and able to write, code, and do math with more accuracy and consistency, it won’t be long before AI makes its way into the workplace as a daily-use tool. In fact, OpenAI itself is betting that a vast majority of workers will have at least part of their jobs automated by GPT models.
In a paper posted to the arXiv preprint server, researchers from OpenAI and the University of Pennsylvania argued that 80 percent of the US workforce could have at least 10 percent of their tasks affected by the introduction of GPTs, the series of popular large language models made by OpenAI. They also found that around 19 percent of workers will see at least 50 percent of their tasks impacted. GPT exposure is greater for higher-income jobs, they wrote in the study, but spans across almost all industries. They argue that GPT models are general-purpose technologies like the steam engine or the printing press.
The researchers used the O*NET database, which is the primary occupation database in the U.S. and lists 1,016 occupations with standardized descriptions, to determine the tasks to measure for each occupation. They then collected both human and GPT-4 generated annotations using a rubric to determine if access to GPT directly or a secondary GPT-powered system would reduce the time required for a human to perform a specific task by at least 50 percent. Higher exposure meant that GPT would reduce the time required to complete the task by at least half while maintaining high-quality work.
“Our findings indicate that the importance of science and critical thinking skills are strongly negatively associated with exposure, suggesting that occupations requiring these skills are less likely to be impacted by current language models,” the researchers wrote. “Conversely, programming and writing skills show a strong positive association with exposure, implying that occupations involving these skills are more susceptible to being influenced by language models.”
The occupations with the highest exposure include mathematicians, tax preparers, writers, web designers, accountants, journalists, and legal secretaries. The occupations with the highest variance, or less likely to be impacted by GPT, include graphic designers, search marketing strategies, and financial managers.
The researchers also list GPT’s overall anticipated impact on different industries, with the greatest impact being on data processing services, information services, publishing industries, and insurance carriers, while the least impact is on food manufacturing, wood product manufacturing, and support activities for agriculture and forestry.
The researchers acknowledge that their study has limitations, due to the fact that the human annotators were familiar with the models’ capabilities and did not belong to some of the occupations measured. Another limitation included that GPT-4 is sensitive to the prompt’s wording and composition, as well as can sometimes make up information, so its outputs are not necessarily the definitive truth. Of course, it should be noted that OpenAI itself produced the work, and as a for-profit company developing AI models it has a high incentive to portray its tools as disrupting industries and automating tasks, which in the end benefits employers.
Still, the report reveals how GPT models will soon be a commonly used tool. Google and Microsoft have already announced that they would be incorporating AI in their office products such as email, documents, and in their search engines. Already, startups are using GPT-4 and its coding abilities to reduce how much they spend on human developers.
“Our analysis indicates that the impacts of LLMs like GPT-4, are likely to be pervasive,” the researchers write. “While LLMs have consistently improved in capabilities over time, their growing economic effect is expected to persist and increase even if we halt the development of new capabilities today.”
OpenAI did not immediately respond to a request for comment.